PR Detailed Description -- Summarized by Claude

Companion: clawpack/geoclaw#745, which adds the num_dim argument and the 1D
write paths. Merge geoclaw#745 first. It defaults to num_dim=2, so it is
inert on its own; this PR landing first against an older geoclaw would
otherwise break every geoclaw setrun. The fallback below covers that case
regardless.

Why this is needed

ClawRunData builds the same TopographyData/DTopoData objects for 1D and
2D runs, but src/1d_classic and src/2d/shallow do not share a Fortran
reader, and the 1D reader was never updated for the per-file preprocessing
block that clawpack/geoclaw#726 introduced. Every examples/1d_classic case
aborts at startup reading topo.data.

Those classes cannot infer the dimension themselves, and
ClawData.__setattr__ rejects unknown attributes, so setting num_dim after
construction is not an option either. Passing it in at construction is the
route already used for the neighbouring objects:

self.add_data(amrclaw.GaugeData(num_dim=num_dim), 'gaugedata')
self.add_data(amrclaw.RegionData(num_dim=num_dim), 'regiondata')
self.add_data(amrclaw.AdjointData(num_dim=num_dim), 'adjointdata')

TopographyData and DTopoData were the odd ones out.

The change

Four call sites needed it — TopographyData and DTopoData, in both the
geoclaw and the dclaw package branches — so rather than repeat a
try/except four times it goes through one helper:

def _make_dim_aware(cls, num_dim):
    try:
        return cls(num_dim=num_dim)
    except TypeError:
        return cls()

dclaw is 2D in practice, but it uses the same geoclaw data classes and
passing num_dim there is correct and harmless.

On the fallback

It exists specifically for mixed clawutil/geoclaw checkouts, which are routine
during development — the 1D breakage was found in exactly that situation, with
several repos on different branches. Without it, a clawutil ahead of geoclaw
turns a TypeError in ClawRunData.__init__ into a failure of every geoclaw
setrun.py, which is a much worse failure than the one being fixed.

It is deliberately temporary and the docstring says so: once the geoclaw
requirement catches up, _make_dim_aware should collapse to a direct call.
Catching TypeError is narrow enough here because neither __init__ raises
TypeError internally — both only call add_attribute.

Verification

Exercised against the geoclaw side of the pair:

• ClawRunData('geoclaw', 1) — topo_data.num_dim == 1 and
  dtopo_data.num_dim == 1.
• ClawRunData('geoclaw', 2) — both 2; 2D data files byte-identical to
  before.
• All six examples/1d_classic cases build and run end to end, which is the
  behaviour this argument exists to enable.
• geoclaw pytest tests/ -m "not remote": 519 passed, 4 skipped, 1 xfailed.
• geoclaw examples/tsunami/chile2010 (2D, exercises both topo and dtopo):
  6 passed.

The fallback path was confirmed by construction rather than by pinning an old
geoclaw: with geoclaw#745 applied, cls(num_dim=...) succeeds and the
except branch is not taken.